@Flashback: Yeah, that's been that way for a long time (games trying to be movies). The gamer has wanted more and more levels of realism. Yet, computers (even dedicated ones like PS3, etc) still have to catch up to the desire. Glad that I don't do that that type of work anymore. I'm pretty happy just doing my scientific programming (Granted, we still have a lot of the same issues... one of the images that I work with is an image stack that is 800,000 x 900,000 pixel x 200 image planes... 1.5 terabyte uncompressed... Trying to 1) build that image and 2) trying to display that image reasonably on a 32-bit machine sucks... then trying to add a high-res detail neuron structure inside of it... lots and lots of coding tricks.)
The issue with the polygon count being so high is the memory usage and the processing power for the algorithms to render the scene correctly. In a lot of cases, you can separate the model into various polygon groups and change their level of detail before you render them, but that computation takes some processing power also...and some fairly nasty math.
Like every other technical field, there is a lot of hidden issues that are tough to resolve and, most people don't encounter them, but when they do ... and they solve them, it's a great feeling.